Memory

Persistent knowledge in CRHQ — how agents remember facts, preferences, and context across sessions.

Memory gives agents persistent knowledge that carries across sessions. Instead of starting fresh every conversation, agents can recall facts, preferences, decisions, and context you've established before.

How Memory Works

When important information comes up in a conversation — your preferences, key decisions, project details, or technical facts — agents can store it as a memory. In future sessions, agents automatically search memory for relevant context.

Memory Types

Memories are categorized by type:

TypePurposeExample
FactObjective information"The production database is PostgreSQL 17 on port 5432"
PreferenceHow you like things done"Always use TypeScript, never plain JavaScript"
DecisionChoices that were made"We chose Stripe over PayPal for payment processing"
ContextBackground information"The client's target audience is enterprise SaaS buyers"

Automatic vs Manual

  • Automatic — Agents can store memories during conversations when they encounter important information
  • Manual — You can create memories directly in Settings → Memory

Tags

Memories can be tagged for organization. Tags help both humans and agents find relevant information:

  • #project-name — Project-specific context
  • #technical — Technical decisions and configurations
  • #client — Client preferences and requirements
  • #process — How things should be done

Memory supports two types of search:

  • Full-text search — Find memories by exact keywords
  • Semantic search — Find memories by meaning (powered by vector embeddings)

Semantic search means an agent searching for "database setup" will also find memories about "PostgreSQL configuration" — even if those exact words weren't used.